Schema Graphs and AI Readability

Reading time: 12 minutes.

{
  "@context": "https://schema.org",
  "@graph": [
    { "@type": "Article",
      "@id": "https://example.com/page/#article",
      "headline": "...",
      "author": { "@id": "https://example.com/#author" },
      "publisher": { "@id": "https://example.com/#organization" },
      "datePublished": "2026-04-25",
      "isPartOf": { "@id": "https://example.com/#website" } },
    { "@type": "Person",
      "@id": "https://example.com/#author", "name": "...", "knowsAbout": [...] },
    { "@type": "Organization",
      "@id": "https://example.com/#organization", "sameAs": [...] },
    { "@type": "WebSite",
      "@id": "https://example.com/#website" },
    { "@type": "BreadcrumbList", "itemListElement": [...] },
    { "@type": "FAQPage", "mainEntity": [...] }
  ]
}

The JSON-LD @graph: One Block, All Entities

Most agency content treats schema as a checklist of separate snippets: an Article block here, an Organization block there, an FAQPage block at the bottom. Each block validates in isolation, but AI engines do not parse them in isolation. They parse the entire JSON-LD payload on a page as a connected graph and weight relationships between entities as evidence of context, authorship, and entity coherence.

The @graph construct in JSON-LD is the schema.org-blessed way to declare those relationships explicitly. One @graph array on a page contains every entity the page is about, every entity the page is by, and every entity the page is part of, all linked through stable @id URLs. The article's author @id resolves to the Person block. The Person's worksFor resolves to the Organization block. The Organization's publisher back-reference closes the loop. Reading the graph, an AI engine sees a coherent network: a named author, working for a named organization, publishing on a website, producing this article, with these citations.

Practically, the @graph block sits in the document head as a single <script type="application/ld+json">. Industry tracking confirms head placement is preferred by Google and processed reliably by AI engines that scrape pages outside the rendered DOM. JSON-LD in the body still works but is more brittle to caching, rendering, and partial-load conditions. The head is the right place; the @graph is the right shape; the discipline is to keep all schema in one place and one structure.

Article + Person + Organization: The Required Triad

The minimum viable @graph for a cornerstone article in this niche is six entities, but the irreducible core is three: Article, Person, and Organization. These three carry the E-E-A-T signal that Google's September 2025 Quality Rater Guidelines update made the dominant evaluation criterion. Stripping any one of them invalidates the trust signal.

The graph closes when these three entities reference each other through stable @id URLs. The Article's author @id equals the Person's @id. The Person's worksFor @id equals the Organization's @id. The Organization's founder can equal the Person's @id for a founder-led brand. Three entities, three back-references, one coherent network.

Matt Griffin notes in client briefings that "ninety percent of agency-produced schema we audit is technically valid in the Schema Markup Validator and structurally incoherent as a graph. The validator confirms each entity in isolation; it does not check whether the graph relationships actually resolve. AI engines do."

FAQPage and the 3.2x Citation Lift

FAQPage schema deserves a section of its own because the citation impact in 2026 is asymmetric. Pages with FAQPage schema show roughly 3.2x higher citation rates in AI Overviews compared to pages without it, even after Google's March 2026 update reduced rich result display for FAQ content on non-primary pages. The rich snippet visibility decreased; the AI citation eligibility did not.

The mechanic is straightforward. AI Overviews and Perplexity look for question-answer pairs to extract directly into their answer responses. A page with FAQPage schema makes those pairs explicit and machine-readable, and the engines reward the explicitness with higher citation probability. The schema is not the cause of the lift in isolation; the lift comes from the engine's ability to confidently lift the question-answer pair without inferring it from prose.

The discipline that makes FAQPage schema work is also straightforward. Every visible FAQ on the page must appear in the FAQPage schema. Every FAQ in the schema must have a visible counterpart. Question text and answer text must match word-for-word between the visible HTML and the JSON-LD. Drift between the two surfaces is a trust violation that downgrades the entire page, not just the FAQ section. The Vector 4 (Embed) work and the Vector 6 schema work share this constraint and have to be done in coordination, not as separate later passes.

The sameAs Array: Connecting to Vector 2

The Organization schema's sameAs array is the single highest-leverage property in the entire schema graph because it carries the cross-engine propagation effect Vector 2 installed. The sameAs array on Organization should include the Wikidata URL, any Wikipedia article in any language, the LinkedIn company page, Crunchbase, X/Twitter, Facebook, Instagram, YouTube, and registry identifiers (Companies House, OpenCorporates, regional business registries).

Each URL in the array is an assertion that the brand at this domain is the same entity as the entity at the linked URL. AI engines cross-validate the assertions during retrieval; confirmed sameAs relationships strengthen the entity graph and feed downstream into Knowledge Panel accuracy, AI Overview citation behaviour, and Perplexity source attribution. The array does not need to be exhaustive; it needs to be accurate, and every URL has to actually resolve and actually represent the same entity.

If your existing pages carry valid-but-incoherent schema, the retrofit pass is high-leverage. A Vector 6 schema audit typically lifts AI citation rates measurably within the next crawl cycle without changing visible content.

Schema Drift: Why FAQ Mismatches Get Penalized

The most common Vector 6 error in production is schema drift, a state where the JSON-LD on a page no longer matches the visible content. The drift typically starts innocently: a writer edits an FAQ answer for clarity but does not update the schema. A developer rolls out a template change that adds new FAQs but does not regenerate the JSON-LD. A CMS migration moves the FAQ block but leaves the original schema in place.

Each instance of drift is a trust signal degradation. Google's machine readers compare visible content to declared schema during indexing; when the two surfaces diverge, the schema is interpreted as either misleading (a violation) or stale (a quality issue). AI engines that consume the same schema for citation decisions register the same divergence and downgrade citation eligibility accordingly. The penalty is not always loud, but it accumulates across a content library and produces site-wide visibility decay over time.

The remediation pattern is automated coherence checking on every deploy. Before any page goes live, the visible question-answer text in the FAQ section gets diffed against the FAQPage schema text; mismatches block the deploy until reconciled. The Article's headline and the visible H1 get cross-checked. The Organization sameAs URLs get pinged for resolution. The discipline is engineering hygiene, not creative judgment.

From Structure to Distribute: The Vector 6 Handoff

Vector 6 is the schema stage; Vector 7 is the distribution stage. The handoff is the way the on-page schema graph connects to off-page citation networks. The Organization sameAs array points outward to the brand's external identity references; Vector 7's work is to grow that array by earning new external mentions in publications and directories that AI engines train on. The schema graph on the brand's domain is the inward expression of the entity; the distribution work is the outward expression. Both are required, and both need each other.

The downstream measurement vectors (Vector 11 Measure, Vector 12 Iterate) read the schema graph indirectly through citation outcomes: when AI Overviews start citing the brand, the citations contain references to the schema metadata (the named author, the published date, the breadcrumb path). The schema is what makes the citations machine-readable in both directions: the engines pull from it during retrieval, and the engines reference it during attribution.

Frequently Asked Questions